Unified Marketing Measurement - Closing the Gap in Marketing Analytics
In 2022 cookie deprecation is real, privacy regulations are strict and marketing is being held more and more accountable for measurable contribution to revenue. With these challenges it has become essential to analyze the full circle impact of the marketing mix while at the same time being able to adjust campaigns during their run time to make the most of the marketing budget. Unified Marketing Measurement solves for these requirements - read further to see how.
Unified Marketing Measurement (UMM) applies statistical models and provides combined insights into both - the overall marketing mix and the customer journey. It enables marketers to evaluate the impact of activities where only aggregated data is available such as print or TV on a user-level customer journey.
But let’s take a step back and look first at the two components of UMM: Marketing Mix Modeling and Multi Touch Attribution. Knowing how they work will help you understand the benefits of UMM and how it enables you to prove marketing ROI.
Unified Marketing Measurement Combines Insights from User-Level and Aggregated Data.
Measure Impact from Offline Channels with Marketing-Mix-Modeling
Marketing Mix Modeling (MMM) is a proven and tested statistical measurement method that has been applied by companies such as Procter & Gamble since the 1980s. It helps quantify the impact of marketing activities on revenue. This information enables marketers to decide where to allocate their budgets strategically to leverage ROI.
MMM collects data on an aggregated level from all kinds of sources both online and offline. You can plug in your TV activities, paid search data, mailings, print campaigns or Impressions from Social Media - all without the need for personal identifiable information (PII). With privacy regulations making data collection even harder nowadays this is one of the biggest advantages of MMM. Marketing analysis can run smoothly without any consent management needed, providing a macro level view on how the marketing mix is working out.
But what does MMM actually do once data is collected? It runs a regression (in our case at Adtriba a Bayesian Time Series Regression) working with input variables, factoring in control variables and predicting an outcome in the form of a target variable. Don’t worry - I hear you! You’re not a statistician - so let’s go through this concept step by step:
This regression analysis works with time-series data - which simply means you’re collecting data together with information about time e.g. sales today, sales yesterday, sales the day before etc resulting in thousands of data points. With some additional effort the algorithm is even able to learn not only how big the impact of an activity is on sales but how it develops over time accounting for the impact from older ads (called adstock effect).
Part of the data is defined as control variables such as price or placement and external factors in the form of weather and seasonality. It is important to control for these factors since they do affect your sales in addition to your marketing activities.
Another part of the data contains summed up marketing activities such as clicks on an ad banner or sent emails. Adstock is included here to make the model even more precise.
These data points are modelled on a target variable which in our case is always some sort of desired output e.g. conversions, sales or revenue.
A very simplified equation for this can look like this ():
© Adtriba 2022
Once the data is plugged into the calculation, a marketer can receive answers to different business questions such as “what is the incremental impact of marketing activities on sales?”
Wait, what does incremental even mean? Incrementality shows how many more sales you can generate with the next additional marketing dollar spent or tv spot run. Naturally not all activities impact sales in a linear way - usually a point of saturation arrives sooner or later when running more TV ads will not get people to buy equally more from your company.
This law of diminishing returns can and should be accounted for by MMM and is incredibly helpful in understanding not only how your current marketing mix is performing but as well where to put your marketing resources next and which channel to explore further.
To sum up MMM helps marketers to understand the incremental impact of marketing activities on sales in a holistic view. This information helps to to make certain and evidence-based decisions to increase overall ROI by prioritizing high ROI activities - all in a privacy-friendly way working with aggregated data.
Multi-Touch-Attribution works with user-level data
Gain Customer Journey Insights with Multi-Touch-Attribution
While MMM works on aggregated data, measures incrementality and helps predict and even optimize sales on a macro level - Multi Touch Attribution (MTA) takes a different approach.
To run an MTA model, data is collected on user-level from digital online sources with GDPR compliant consent resulting in millions of data points. Data from each touch-point along the customer journey is included e.g. social ad clicks, paid search clicks, onsite events such as add-to-carts and alike. MTA differentiates between converting and non-converting customer journeys and helps understand the impact a single touchpoint has on the conversion probability - the result is a calculated attribution weight.
Simply said: MTA helps you understand how likely a customer is to convert when you show them a Facebook ad vs when you don’t. And how this likelihood changes with respect to other touchpoints he or she already had with your company.
Multi Touch Attribution runs in contrast to MMM not on a regression but on a classification model which is a different statistical method to analyze data. It does not factor in seasonality or other external factors nor does it consider offline sources. The result is a micro view on marketing activities and their impact on sales.
You will see later how UMM helps to bridge this gap and bring the effects between MMM and MTA to light.
For now let’s focus on the classification applied in MTA which compares converting and non-converting customer journeys. The MTA algorithm deduces through subtraction how much each particular touchpoint impacts the conversion.
Imagine the following scenario: a person who has seen your Facebook ad searches for your brand on google, clicks on a paid search ad, browses the site, adds an item to cart, leaves - and returns after clicking on a retargeting ad to finally convert. Another person might go the exact same route but without the retargeting ad. By comparing these two scenarios MTA is able to specify what exact impact the retargeting ad had on the conversion.
MTA applies Machine Learning to recognize recurring patterns in sequential events e.g. the order of touch-points in a journey. The particular algorithm is called Long-Short-Term-Memory (LSTM) which is not only applied in marketing but in all sorts of machine-learning challenges such as text suggestions, translation services and other.
The general approach of LSTM is to feed information into sequential loops and allow it to persist. This takes into account events that took place a long time ago in combination with recent short term events (hence the name). By bridging the time gap between touchpoints in a journey the model is working precisely applying if-then-rules. Those determine for the algorithm how long to “remember” a certain information e.g. a click on an ad.
Further LSTM solves for a typical marketing complexity of multiple layers to be analysed. Imagine running 20 campaigns across 5 channels in 5 languages - this already results in 1000 different scenarios to consider.
But how do we know whether or not the results are valid? To make sure the model is trustworthy a procedure of a hold-out sample is applied. Only 70% of the data points are fed into the LSTM model, the other 30% are put to the side (held out). The model developed and trained on the 70% and is then applied on the 30% of the hold out data. The resulting model predictions are then compared to the true and raw reality data to see how closely they match.
Once this validation is done the attribution weights which show the impact of each touchpoint on a conversion are calculated by subtraction. The result is a weight factor between 0 - 1 indicating that a certain touchpoint - let’s say a retargeting banner - impacts conversions by x%.
As you can see MTA is incredibly helpful with understanding the impact of a single activity on a customer journey providing you with a detailed micro view on your activities. However, it does not consider seasonality effects, pricing or other external factors. And it doesn’t take brand effects which impact baseline sales into account. Nor does it include the impact of offline sources.
In an ideal marketing analysis world however, you would need to understand both the macro and the micro view on your activities to make solid business decisions. This is where UMM helps out by combining MMM and MTA into a holistic view.
Unified Marketing Measurement lives on the intersect between MMM and MTA
How Unified Marketing Measurement combines Insights from MMM and MTA
Unified Marketing Measurement (UMM) lives on the intersection between Marketing Mixed Modeling and Multi Touch Attribution combining insights from aggregated and user-level data. This enables marketers to answer both activity planning and budget allocation questions.
What is special about UMM is that it receives input from aggregated data collected through MMM such as TV marketing spend on a particular day and calculates the impact on a certain customer journey as part of the MTA (see image below © Adtriba 2022). By this UMM helps to bridge the gap between data that is only available on aggregated level with user-level data.
A simplified example might look like this: A print campaign for luxury design stationery is currently running in magazines complemented by a paid search campaign.
The print campaign can be analyzed as part of MMM on an aggregated level to help understand how it affected the overall sales during and after the campaign’s lifespan. The paid search campaign can be analyzed on a user-level as part of the MTA to reveal the exact impact on customer journeys. Combining these insights in the UMM will take this a step further - we can then understand the impact of the print campaign on a customer journey because we can calculate it’s impact on paid search.
Knowing this would further help assess the effectiveness of the print campaign and help to decide, whether or not to do it again and whether to combine it with other complementary activities such as paid search, mailing or social media campaigns.
So how does UMM's magic work here? First a marketing mix model is developed to analyze how marketing activities, including print campaigns, impact paid search clicks. The resulting model considers all external factors such as seasonality or any global pandemics taking place while accounting for adstock effects from previous marketing activities.
The output from the MMM macro analysis in the form of a proportional factor - let’s say it’s 0.3 (how much impact print had on paid search on a given day) - is plugged into the MTA micro analysis. Now the previously aggregated data about the results from the print campaign can be treated as a touchpoint in customer journeys.
With plugged in output from MMM regression the MTA runs the LSTM loops and reveals that in particular converting customer journey the paid search click affects sales by 0.5.
We’re not done yet, however. This insights are now taken back to the MMM multiplying the 0.3 proportional factor from before with the newly calculated 0.5 revealing the true impact of the print campaign on this conversion being 0.3*0.5=0.15.
Now this is fairly detailed and don’t you worry - as a marketer using attribution software to understand the impact of your activities you don’t have to look at single weights to make sense of the big picture - a software such as Adtriba provides you with easy to grasp and actionable insights. Nevertheless, it is important to understand how it all comes into place so you can be confident in your decision making.
And speaking of confidence - how are we so confident that our algorithm is correct supplying you with robust and precise information about your marketing ROI? We ensure the validity of our models using statistical validation measures such as R-squared, Mean Absolute Percentage Error (MAPE) and Area Under the Curve (AUC) calculated with each model to make sure you receive only significant results. These methods show the difference (or error) between the estimated results by the model and the real world data and indicate whether this error is small enough so the results count as valid.
And if that is not confident enough yet - we further enhance our algorithms with a hierarchy of models where we consider that external factors (e.g. seasonality), on-site events (e.g. add to cart) and marketing activities (social media campaigns) do not only impact sales directly, but impact each other creating a complex interdependent system (see image below © Adtriba 2022).
Why should Marketers use a Unified Marketing Measurement Solution?
As a marketer the need for evidence based decisions is increasing and so is the complexity of measurement. It is no longer sufficient to try and make sense of single activities or siloed parts of the customer journey. To make confident and evidence-based decisions it takes a comprehensive holistic approach to measurement. This enables you to make the most of your marketing budget tactically, strategically and operationally.
While Multi Touch Attribution and Marketing Mixed Modeling are well established analytical frameworks, they both come to certain limits with regards to data and effects they are able to consider. MTA does incredible work on micro-level analysing touch-points in customer journeys. But it lacks a bird’s eye view of external effects and it does not consider offline sources. MMM can account for exactly that with a macro-level view of calculating brand effects, seasonality and the effect of whole marketing campaigns on total sales.
Unified Marketing Measurement combines the two methodologies and provides a fully integrated view. By this it is showing the impact of offline sources - where only aggregated data is available - on a user level in a customer journey.
Using a Unified Marketing Measurement solution enables you to measure the exact impact of your cross-channel marketing mix, understand your baseline sales, and see accurate incrementality of your activities. With these insights you are well equipped to leverage ROI while keeping budgets constant.